Message boards :
Number crunching :
Average Credit Decreasing?
Message board moderation
Previous · 1 . . . 3 · 4 · 5 · 6 · 7 · 8 · 9 . . . 32 · Next
Author | Message |
---|---|
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13854 Credit: 208,696,464 RAC: 304 |
All I can say Richard is that that graph isn't representing what I'm seeing at all here. If you check out the graph below that one you can see whether credit is rising, falling or treading water- and in your case it's as you said- plummeting. In my case it peaked on the 9/04/216 & has been dropping since then (although if you look as the BS-RAC you get something very different). Grant Darwin NT |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14679 Credit: 200,643,578 RAC: 874 |
[Edit:] @Richard: noticed you have a 970 in with a 750ti in two machines ? both those are likely to claim on the high side for 750ti tasks... (peak_flops_from_wrong_device X elapsed) I did declare the 970s in my original statement of evidence, m'lud... But I hadn't picked up from the previous discussion of code that mis-matched GPUs would have that effect (perhaps because I never ran mis-matched GPUs until January). If that's the case, there's no way of applying rigorous mathematics to credit until the server has proper, individual, records of each separate GPU. On another tack, I'me seeing a remarkably low proportion of shorties in the current Arecibo mix - that's a big plus for my 750s. |
William Send message Joined: 14 Feb 13 Posts: 2037 Credit: 17,689,662 RAC: 0 |
[Edit:] @Richard: noticed you have a 970 in with a 750ti in two machines ? both those are likely to claim on the high side for 750ti tasks... (peak_flops_from_wrong_device X elapsed) please don't mention maths and CreditNew in the same sentence... Next time I need a RNG I'm just going to use credit. A person who won't read has no advantage over one who can't read. (Mark Twain) |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14679 Credit: 200,643,578 RAC: 874 |
please don't mention maths and CreditNew in the same sentence... lol. But to be serious, I do think that we should be able to mention all three of Maths, Science, and Credit in the same BOINC sentence. It worries me greatly at projects like CPDN, which also has constant struggles to fit within the BOINC credit framework. It's getting better now, but in the early days of that project, it was up against some fairly vociferous opposition from climate change deniers. Every time they had a credit screw-up, I expected posts like "if you can't even get the maths right for your own credits, why should we believe a word you say about the weather?". And I wouldn't have had an answer. No, now BOINC has created the credit monster, I think it has a responsibility to calculate it with the same rigour as the projects calculate their science. And we have a responsibility to report the problems as fully and accurately as we can. (before anyone asks, CPDN doesn't use, and has never used, CreditNew) |
William Send message Joined: 14 Feb 13 Posts: 2037 Credit: 17,689,662 RAC: 0 |
No, now BOINC has created the credit monster, I think it has a responsibility to calculate it with the same rigour as the projects calculate their science. And we have a responsibility to report the problems as fully and accurately as we can. No objections, your honour. But credit should have a sound mathematical basis and a faultless computational algorithm implementation. And CreditNew hasn't. Not the latter and I'm still not convinced about the former. Looking at credit in this detail is like looking at a house that is falling down, has gaps all over the planks and a leaking roof, when the architect used a twig for a ruler and the builders used cardboard for wood... A person who won't read has no advantage over one who can't read. (Mark Twain) |
Ulrich Metzner Send message Joined: 3 Jul 02 Posts: 1256 Credit: 13,565,513 RAC: 13 |
(...) That - and every serious engineer or architect would only take one choice: Pull it! Aloha, Uli |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14679 Credit: 200,643,578 RAC: 874 |
(...) But then everyone who crunches for credit would be homeless! |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
I'm fairly convinced that for the purposes of estimation/scheduling, +/- 10% would be good enough for most under steady state, along with faster/better adaptation to work or running condition changes, without overshoot or oscillations. Also more than achievable without having to invent any new special technologies. Keeping the adaptive capability is important for situations such as these new telescopes coming online (or new GPUs, apps, etc). In these respects I think it works as 'proof of concept', but shows quite a few design limitations and implementation problems, that if budget and resources were allocated to address probably we wouldn't need these discussions everytime something or another changes. Part of the problems I;ve watched people wrestle with while walking the associated Boinc code, is that there isn;t a clear distinction between the design intent, and the implementation, so dropping in refinements for so much as a test run is bigger than a trivial job. To be fair on the system in that respect, it shows the same problems with flexibility, and creaking under the demands for change, as the applications are starting to. In the case of the applications, it';s realtively easy to start afresh, and there are existing applications to compare against for correctness and other metrics. For the server code and specifically credit as a measurer/predictor of work, afaik we only have the cobblestone scale value (#operations on a psuedo-currency scale). It's true that number can often have a reasoable value per work unit (as it does here for the most part), however the other part, efficiencies on the different levels: Project-?App-?version/planclass->host->device are not localised to each of those levels. Completing that estimate chain, for example figuring out the rough overall efficiency of a particular host-app-version-platfor-planclass, and percolating that up to summary statisics would solve a lot of measuremen, prediction, monitoring and control problems. Preferably something better than averages which are wasting database resources, not responsive enough to change, and susceptible to noise+skew. There's nothing particularly new/innoivative/complex about that, just that such systems require resources to implement properly, along with clear design goals (such as please do steady state estimates to +/- 10% or better). For those parts I think the full potential uses of a real and complete implementation would be well worth it ---> If Boinc wasn't out of dough. "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
William Send message Joined: 14 Feb 13 Posts: 2037 Credit: 17,689,662 RAC: 0 |
ok, so what we really need is somebody with a bucketful of cash, who says 'here have that - but only if you make sure credit works' ? A person who won't read has no advantage over one who can't read. (Mark Twain) |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
ok, so what we really need is somebody with a bucketful of cash, who says 'here have that - but only if you make sure credit works' ? And change the godawful name 'CreditNew' to '[Throughput-]Estimator', or something. It makes people think it's about credits, so it never gets fixed. "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
Dimly Lit Lightbulb 😀 Send message Joined: 30 Aug 08 Posts: 15399 Credit: 7,423,413 RAC: 1 |
For me the earlie Arecibo ~97-104 credit , 0.41 AR, seem to be around the 77-84 Cridt range, though clearly still settling down. Sadly the coming telescope soup is likey to make it much less simple, so I've pretty much resigned myself into buying a VW combivan and touring the world. Jason and the Algorithms are going on tour? *books tickets* :) Member of the People Encouraging Niceness In Society club. |
Jord Send message Joined: 9 Jun 99 Posts: 15184 Credit: 4,362,181 RAC: 3 |
No, now BOINC has created the credit monster, I think it has a responsibility to calculate it with the same rigour as the projects calculate their science. How about setting up a project that recalculates everyone's credit gained through CN and any other credit generation and adjusts it all down to one common numeral? We can call it "[Throughput-]Numerator-Denominator-Estimator". A bit like WUProp, but for credit. :) |
OzzFan Send message Joined: 9 Apr 02 Posts: 15691 Credit: 84,761,841 RAC: 28 |
No, now BOINC has created the credit monster, I think it has a responsibility to calculate it with the same rigour as the projects calculate their science. I think I like the name "Fairly UnderCut Kredit Estimator Denominator". Now if only we could come up with a good acronym for it. |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14679 Credit: 200,643,578 RAC: 874 |
How about setting up a project that recalculates everyone's credit gained through CN and any other credit generation and adjusts it all down to one common numeral? We can call it "[Throughput-]Numerator-Denominator-Estimator". Actually, we have one of those already. Project granted credit comparison Check the units those numbers are expressed in, very carefully. |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
Maybe we should go back to the something like set amount of credit for VALIDATED work. That would resolve a lot of things. |
Wiggo Send message Joined: 24 Jan 00 Posts: 36779 Credit: 261,360,520 RAC: 489 |
Maybe we should go back to the something like set amount of credit for VALIDATED work. That would resolve a lot of things. Yes, at least that system worked with any degradation. Cheers. |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13854 Credit: 208,696,464 RAC: 304 |
Maybe we should go back to the something like set amount of credit for VALIDATED work. That would resolve a lot of things. I think the emphasis there should be on set amount of credit; as things are now you still only get credit for validated work. It's just that the amount varies- generally downwards. Grant Darwin NT |
Brent Norman Send message Joined: 1 Dec 99 Posts: 2786 Credit: 685,657,289 RAC: 835 |
I have to disagree Grant, only Credits for Valid work is a bad thing? I could crank up my clock and spit out 14,000 invalids and increase my RAC in that case. |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13854 Credit: 208,696,464 RAC: 304 |
I have to disagree Grant, only Credits for Valid work is a bad thing? I'm not saying it's a bad thing. The fact is that is what we already have, it doesn't need changing so why emphasize that and ignore the very thing that is being discussed here- the random nature of credit being granted? Hence my emphasis on a fixed a mount of credit for a given amount of work. That's what's broken, hence the emphasis. Credit only for valid work isn't broken, so why emphasise it? Grant Darwin NT |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
I emphasised that section, as in the past, credit was granted for work done, not validated work. So yes, people cranked up their GPUs and returned tons of invalid work. A fixed amount would help but the work needs the validation so that we don't have a repeat. But you are right, we are off topic... |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.